Python functions.explode函数代码示例

    def test_explode(self):
        from pyspark.sql.functions import explode, explode_outer, posexplode_outer
        d = [
            Row(a=1, intlist=[1, 2, 3], mapfield={"a": "b"}),
            Row(a=1, intlist=[], mapfield={}),
            Row(a=1, intlist=None, mapfield=None),
        ]
        rdd = self.sc.parallelize(d)
        data = self.spark.createDataFrame(rdd)

        result = data.select(explode(data.intlist).alias("a")).select("a").collect()
        self.assertEqual(result[0][0], 1)
        self.assertEqual(result[1][0], 2)
        self.assertEqual(result[2][0], 3)

        result = data.select(explode(data.mapfield).alias("a", "b")).select("a", "b").collect()
        self.assertEqual(result[0][0], "a")
        self.assertEqual(result[0][1], "b")

        result = [tuple(x) for x in data.select(posexplode_outer("intlist")).collect()]
        self.assertEqual(result, [(0, 1), (1, 2), (2, 3), (None, None), (None, None)])

        result = [tuple(x) for x in data.select(posexplode_outer("mapfield")).collect()]
        self.assertEqual(result, [(0, 'a', 'b'), (None, None, None), (None, None, None)])

        result = [x[0] for x in data.select(explode_outer("intlist")).collect()]
        self.assertEqual(result, [1, 2, 3, None, None])

        result = [tuple(x) for x in data.select(explode_outer("mapfield")).collect()]
        self.assertEqual(result, [('a', 'b'), (None, None), (None, None)])

def process(rdd):
  print(">>>> BEGIN CASS")
  wonbids = getSqlContextInstance(rdd.context).createDataFrame(rdd) 
  wonbids.registerTempTable("wonbids")
  wonbids.write.format("org.apache.spark.sql.cassandra").\
           options(keyspace="text_bids", table="bidswon").\
           save(mode="append")
  #sqlContext.cacheTable('wonbids')
  # wonbids.show()
 
  symptoms = wonbids.select(wonbids.id,wonbids.created_utc,explode(wonbids.symptomtags).alias('symptom'))
  symptoms.registerTempTable("symptoms")
  symptoms.write.format("org.apache.spark.sql.cassandra").\
         options(keyspace="text_bids", table="symptoms").\
         save(mode="append")
  # symptoms.show()

  conditions = wonbids.select(wonbids.id,wonbids.created_utc,explode(wonbids.conditiontags).alias('condition'))
  conditions.registerTempTable("conditions")
  conditions.write.format("org.apache.spark.sql.cassandra").\
         options(keyspace="text_bids", table="conditions").\
         save(mode="append")
  # conditions.show()
  
  # send back to master to process
  for w in wonbids.collect():
    event.Event('toES', {'id':w.id,'pharmatag':w.pharmatag,'price':w.price,'created_utc':w.created_utc,'symptomtags':w.symptomtags,'conditiontags':w.conditiontags})
  print(">>>> END CASS")

    def test_udf_in_generate(self):
        from pyspark.sql.functions import udf, explode
        df = self.spark.range(5)
        f = udf(lambda x: list(range(x)), ArrayType(LongType()))
        row = df.select(explode(f(*df))).groupBy().sum().first()
        self.assertEqual(row[0], 10)

        df = self.spark.range(3)
        res = df.select("id", explode(f(df.id))).collect()
        self.assertEqual(res[0][0], 1)
        self.assertEqual(res[0][1], 0)
        self.assertEqual(res[1][0], 2)
        self.assertEqual(res[1][1], 0)
        self.assertEqual(res[2][0], 2)
        self.assertEqual(res[2][1], 1)

        range_udf = udf(lambda value: list(range(value - 1, value + 1)), ArrayType(IntegerType()))
        res = df.select("id", explode(range_udf(df.id))).collect()
        self.assertEqual(res[0][0], 0)
        self.assertEqual(res[0][1], -1)
        self.assertEqual(res[1][0], 0)
        self.assertEqual(res[1][1], 0)
        self.assertEqual(res[2][0], 1)
        self.assertEqual(res[2][1], 0)
        self.assertEqual(res[3][0], 1)
        self.assertEqual(res[3][1], 1)

    def test_explode(self):
        from pyspark.sql.functions import explode
        d = [Row(a=1, intlist=[1, 2, 3], mapfield={"a": "b"})]
        rdd = self.sc.parallelize(d)
        data = self.sqlCtx.createDataFrame(rdd)

        result = data.select(explode(data.intlist).alias("a")).select("a").collect()
        self.assertEqual(result[0][0], 1)
        self.assertEqual(result[1][0], 2)
        self.assertEqual(result[2][0], 3)

        result = data.select(explode(data.mapfield).alias("a", "b")).select("a", "b").collect()
        self.assertEqual(result[0][0], "a")
        self.assertEqual(result[0][1], "b")

 def data(self):
     from pyspark.sql.functions import array, explode, col, lit
     return self.spark.range(10).toDF('id') \
         .withColumn("vs", array([lit(i * 1.0) + col('id') for i in range(20, 30)])) \
         .withColumn("v", explode(col('vs'))) \
         .drop('vs') \
         .withColumn('w', lit(1.0))

	def create_one_hot_dict(self,input_df):
		"""Creates a one-hot-encoder dictionary based on the input data.

		Args:
			input_df (DataFrame with 'features' column): A DataFrame where each row contains a list of
				(featureID, value) tuples.

		Returns:
			dict: A dictionary where the keys are (featureID, value) tuples and map to values that are
				unique integers.
		"""
		distinct_feats = input_df.select(explode(input_df.features)).distinct()
		#print distinct_feats.take(1)[0]
		return distinct_feats.rdd.map(lambda x: tuple(x[0])).zipWithIndex().collectAsMap()

def get_average_expected_change(timeframe, partner, purchase):
    """
    returns the average expected change of day 1 to day 7
    input: a period (string), a partner (string), a spark dataframe
    output: a dictionary with key=day and value=average_expected_change
    """
    keys = [1,2,3,4,5,6,7]
    result = {}
    result = {key: 0 for key in keys}
    timeframe_is = get_date(timeframe)
    unlisted = purchase.select(explode(purchase.prediction.days).alias("test")).collect()
    # print 'length of unlisted:', len(unlisted)
    total = len(unlisted) / len(keys)
    # print total
    for i in range(len(unlisted)):
        result[i%7+1] += unlisted[i][0].expected_change / total
    return result

# MAGIC 
# MAGIC Before we can use the `wordcount()` function, we have to address two issues with the format of the DataFrame:
# MAGIC   + The first issue is that  that we need to split each line by its spaces.
# MAGIC   + The second issue is we need to filter out empty lines or words.
# MAGIC 
# MAGIC Apply a transformation that will split each 'sentence' in the DataFrame by its spaces, and then transform from a DataFrame that contains lists of words into a DataFrame with each word in its own row.  To accomplish these two tasks you can use the `split` and `explode` functions found in [pyspark.sql.functions](http://spark.apache.org/docs/latest/api/python/pyspark.sql.html#pyspark.sql.functions).
# MAGIC 
# MAGIC Once you have a DataFrame with one word per row you can apply the [DataFrame operation `where`](http://spark.apache.org/docs/latest/api/python/pyspark.sql.html#pyspark.sql.DataFrame.where) to remove the rows that contain ''.
# MAGIC 
# MAGIC > Note that `shakeWordsDF` should be a DataFrame with one column named `word`.

# COMMAND ----------

# TODO: Replace <FILL IN> with appropriate code
from pyspark.sql.functions import split, explode
shakeWordsDF = (shakespeareDF.select(explode(split(shakespeareDF[0],"\s+")).alias("word"))).where("length(word) > 0")

shakeWordsDF.show()
shakeWordsDFCount = shakeWordsDF.count()
print shakeWordsDFCount

# COMMAND ----------

# TEST Remove empty elements (4d)
Test.assertEquals(shakeWordsDF.count(), 882996, 'incorrect value for shakeWordCount')
Test.assertEquals(shakeWordsDF.columns, ['word'], "shakeWordsDF should only contain the Column 'word'")

# COMMAND ----------

# MAGIC %md
# MAGIC ** (4e) Count the words **

# drop() is like the opposite of select(): Instead of selecting specific columns from a DataFrame, it drops a specifed column from a DataFrame
dataDF.drop('occupation').drop('age').show()

# the sample() transformation returns a new DataFrame with a random sample
sampledDF = dataDF.sample(withReplacement=False, fraction=0.10)
print sampledDF.count()
sampledDF.show()

# split() and explode() transformations
from pyspark.sql.functions import split, explode

shakeWordsSplit = (shakespeareDF
                .select(split(shakespeareDF.word,' ').alias('word'))) # here split(DF,' ') splits the sentence at a space and returns each word in a single row
				
shakeWordsExplode = (shakeWordsSplit
                    .select(explode(shakeWordsSplit.word).alias('word'))) # explode() Returns a new row for each element in the given array
					
shakeWordsDF = shakeWordsExplode.filter(shakeWordsExplode.word != '') # removes all the blanks

shakeWordsDF.show()
shakeWordsDFCount = shakeWordsDF.count()
print shakeWordsDFCount

###############################################################
###															###
###															###
###   						GROUP BY						###
###															###
###															###
###############################################################

from lib.pos_tags import PosTags
from lib.chunks import Chunks

t = Tokens()
p = PosTags()
c = Chunks()

c.train(c.load_training_data("../data/chunker_training_50_fixed.json"))


def pipeline(s):
    """
    Given a string, return a list of relations
    """
    return c.assemble(c.tag(p.tag(t.tokenize(s))))


pipeline_udf = sql.udf(pipeline, types.ArrayType(types.MapType(types.StringType(), types.StringType())))


phrases = (
    notes.withColumn("phrases", pipeline_udf(notes["document"]))
    .select(sql.explode(sql.col("phrases")).alias("text"))
    .filter(sql.col("text")["tag"] == "NP")
    .select(sql.lower(sql.col("text")["phrase"]).alias("phrase"))
    .groupBy(sql.col("phrase"))
    .count()
)

phrases.write.parquet("../data/idigbio_phrases.parquet")

    def rdd_to_recordstore(rdd_transform_context_rdd):

        if rdd_transform_context_rdd.isEmpty():
            MonMetricsKafkaProcessor.log_debug(
                "rdd_to_recordstore: nothing to process...")
        else:

            sql_context = SQLContext(rdd_transform_context_rdd.context)
            data_driven_specs_repo = DataDrivenSpecsRepoFactory.\
                get_data_driven_specs_repo()
            pre_transform_specs_df = data_driven_specs_repo.\
                get_data_driven_specs(
                    sql_context=sql_context,
                    data_driven_spec_type=DataDrivenSpecsRepo.
                    pre_transform_specs_type)

            #
            # extract second column containing raw metric data
            #
            raw_mon_metrics = rdd_transform_context_rdd.map(
                lambda nt: nt.rdd_info[1])

            #
            # convert raw metric data rdd to dataframe rdd
            #
            raw_mon_metrics_df = \
                MonMetricUtils.create_mon_metrics_df_from_json_rdd(
                    sql_context,
                    raw_mon_metrics)

            #
            # filter out unwanted metrics and keep metrics we are interested in
            #
            cond = [
                raw_mon_metrics_df.metric.name ==
                pre_transform_specs_df.event_type]
            filtered_metrics_df = raw_mon_metrics_df.join(
                pre_transform_specs_df, cond)

            #
            # validate filtered metrics to check if required fields
            # are present and not empty
            # In order to be able to apply filter function had to convert
            # data frame rdd to normal rdd. After validation the rdd is
            # converted back to dataframe rdd
            #
            # FIXME: find a way to apply filter function on dataframe rdd data
            validated_mon_metrics_rdd = filtered_metrics_df.rdd.filter(
                MonMetricsKafkaProcessor._validate_raw_mon_metrics)
            validated_mon_metrics_df = sql_context.createDataFrame(
                validated_mon_metrics_rdd, filtered_metrics_df.schema)

            #
            # record generator
            # generate a new intermediate metric record if a given metric
            # metric_id_list, in pre_transform_specs table has several
            # intermediate metrics defined.
            # intermediate metrics are used as a convenient way to
            # process (aggregated) metric in mutiple ways by making a copy
            # of the source data for each processing
            #
            gen_mon_metrics_df = validated_mon_metrics_df.select(
                validated_mon_metrics_df.meta,
                validated_mon_metrics_df.metric,
                validated_mon_metrics_df.event_processing_params,
                validated_mon_metrics_df.event_type,
                explode(validated_mon_metrics_df.metric_id_list).alias(
                    "this_metric_id"),
                validated_mon_metrics_df.service_id)

            #
            # transform metrics data to record_store format
            # record store format is the common format which will serve as
            # source to aggregation processing.
            # converting the metric to common standard format helps in writing
            # generic aggregation routines driven by configuration parameters
            #  and can be reused
            #
            record_store_df = gen_mon_metrics_df.select(
                (gen_mon_metrics_df.metric.timestamp / 1000).alias(
                    "event_timestamp_unix"),
                from_unixtime(
                    gen_mon_metrics_df.metric.timestamp / 1000).alias(
                    "event_timestamp_string"),
                gen_mon_metrics_df.event_type.alias("event_type"),
                gen_mon_metrics_df.event_type.alias("event_quantity_name"),
                (gen_mon_metrics_df.metric.value / 1.0).alias(
                    "event_quantity"),
                when(gen_mon_metrics_df.metric.dimensions.state != '',
                     gen_mon_metrics_df.metric.dimensions.state).otherwise(
                    'NA').alias("event_status"),
                lit('1.0').alias('event_version'),
                lit('metrics').alias("record_type"),

                # resource_uuid
                when(gen_mon_metrics_df.metric.dimensions.instanceId != '',
                     gen_mon_metrics_df.metric.dimensions.instanceId).when(
                    gen_mon_metrics_df.metric.dimensions.resource_id != '',
                    gen_mon_metrics_df.metric.dimensions.resource_id).
                otherwise('NA').alias("resource_uuid"),
#.........这里部分代码省略.........

#linesOut.saveAsTextFile("hdfs:///tmp/fact_icd9_encounter_08242016_supP2_rdd_.csv")

# linesOut.saveAsTextFile('hdfs:///tmp/fact_icd9_encounter_08242016_supP2_rdd.csv')

linesOut.saveAsTextFile("/Users/jayurbain/Dropbox/MCW/fact_icd9_encounter_08242016_supP2_rdd.txt")

linesOut.reduce( lambda k,v: (k))

################

from pyspark.sql import Row
from pyspark.sql.functions import explode

df = sqlContext.createDataFrame([Row(a=1, b=[1,2,3],c=[7,8,9]), Row(a=2, b=[4,5,6],c=[10,11,12])])
df1 = df.select(df.a,explode(df.b).alias("b"),df.c)
df2 = df1.select(df1.a,df1.b,explode(df1.c).alias("c"))



###################################

# fpgrowth example
itemsets = parts.map(lambda p: ( p[3].strip().split(',') ) )
itemsets.getNumPartitions()
model_fp = FPGrowth.train(itemsets, minSupport=0.005, numPartitions=10)
result = model_fp.freqItemsets().collect()
for i in sorted(result, key=operator.itemgetter(1), reverse=True):
    print '(', ', '.join(i.items), ')', 'freq=', str(i.freq)

# MAGIC 
# MAGIC Before we can use the `wordcount()` function, we have to address two issues with the format of the DataFrame:
# MAGIC   + The first issue is that  that we need to split each line by its spaces.
# MAGIC   + The second issue is we need to filter out empty lines or words.
# MAGIC 
# MAGIC Apply a transformation that will split each 'sentence' in the DataFrame by its spaces, and then transform from a DataFrame that contains lists of words into a DataFrame with each word in its own row.  To accomplish these two tasks you can use the `split` and `explode` functions found in [pyspark.sql.functions](http://spark.apache.org/docs/latest/api/python/pyspark.sql.html#pyspark.sql.functions).
# MAGIC 
# MAGIC Once you have a DataFrame with one word per row you can apply the [DataFrame operation `where`](http://spark.apache.org/docs/latest/api/python/pyspark.sql.html#pyspark.sql.DataFrame.where) to remove the rows that contain ''.
# MAGIC 
# MAGIC > Note that `shakeWordsDF` should be a DataFrame with one column named `word`.

# COMMAND ----------

# TODO: Replace <FILL IN> with appropriate code
from pyspark.sql.functions import split, explode
shakeWordsDF = (shakespeareDF.select(explode(split(col('sentence'),' ')).alias('word')))

shakeWordsDF = shakeWordsDF.filter(col('word') != '')

shakeWordsDF.show()
shakeWordsDFCount = shakeWordsDF.count()
print shakeWordsDFCount

# COMMAND ----------

# TEST Remove empty elements (4d)
Test.assertEquals(shakeWordsDF.count(), 882996, 'incorrect value for shakeWordCount')
Test.assertEquals(shakeWordsDF.columns, ['word'], "shakeWordsDF should only contain the Column 'word'")

# COMMAND ----------

 def data(self):
     return self.spark.range(10).toDF('id') \
         .withColumn("vs", array([lit(i) for i in range(20, 30)])) \
         .withColumn("v", explode(col('vs'))).drop('vs')

        casSession = cascluster.connect('test2')
        for aggItem in agg:
            if aggItem[0] != "":
                    casSession.execute('INSERT INTO author_pub_simu (author, pub) VALUES (%s, %s)', (str(aggItem[0]), str(aggItem[1])))
        casSession.shutdown()
        cascluster.shutdown()

sc = SparkContext("spark://ip-172-31-2-40:7077", "2016_test")
sqlContext = SQLContext(sc)

# read in data from HDFS and select columns
df1 = sqlContext.read.json("hdfs://ec2-52-34-128-244.us-west-2.compute.amazonaws.com:9000//simulated/fake_data_p1*.json").dropna()
df_sel = df1.select('recid', 'authors','co-authors','references', 'creation_date').withColumnRenamed('co-authors', 'co_authors').persist(StorageLevel.MEMORY_AND_DISK)

# explode references list and group by citation id to calcualte the number of times that one publication has been cited
df_references = df_sel.select('recid', explode('references')).withColumnRenamed('_c0','cited_id').groupBy('cited_id').count().withColumnRenamed('count','num_cited')

# combine author and co-author list to generate a total list of authors and convert rdd into dataframe
rdd_authors = df_sel.rdd.map(lambda x:{'recid':x.recid, 'authors': append_author(x.authors, x.co_authors), 'creation_date': fetch_year(x.creation_date)})
df_authors = sqlContext.createDataFrame(rdd_authors)

# join citation and author dataframes
df_join = df_references.join(df_authors, df_references.cited_id == df_authors.recid, 'inner').drop(df_authors.recid)

# explode author and save to Cassandra database
df_explode_author = df_join.select('cited_id', 'num_cited', explode('authors'), 'creation_date').withColumnRenamed('_c0', 'author')
df_explode_author.persist(StorageLevel.MEMORY_AND_DISK)
df_sel.unpersist()
df_explode_author.rdd.foreachPartition(aggToCassandra2)

# combine each author publication list,  group by author and calculate H-index for each author